Last week, we announced the immediate availability of Indix Standardized Attributes and Indix Product Descriptions (in partnership with Narrative Science). In a previous post this week, we discussed how we attacked the scale problem of standardizing product attributes to make sense of the crazy product information “soup” that’s found in the wild. In this post, we’ll look at how standardized attributes become product descriptions.
Unless you’ve been living under a rock, you know that Google loves unique content. The better and more relevant your content, the better you’ll rank, which should lead to traffic and conversions (read: $$$). Unique product descriptions come in at #6 in our 9 steps to perfect product page SEO list, and good product descriptions have been shown to increase conversion rates by over 30% (we’ve seen 30%, 78%, and even over 100% in our research).
That said, product descriptions can be difficult to produce. Each description takes a human product description writer approximately an hour to research and produce. This is fine if you’re a boutique ecommerce site with only a thousand or two products, but what happens when you have hundreds of thousands or millions of products?
Looking at the monetary investment alone, producing product descriptions at scale is daunting at best. If you’re trying to hire and use US-based description writers, you’ll shell out approximately $14 per description. If you hire and use offshore talent, your price tag will be closer to $5 per description. For your first few tens of thousands of products (your “head products”), this seems reasonable. If you need 10,000 descriptions, you’ll pay $140,000 for US talent or $50,000 for offshore. It’s not a small price tag, but it’s not undoable.
What happens when you get past your head products, however? Products that haven’t been performing thus far or that you just don’t know enough about to want to invest in them make that price tag start looking steep. Outsourcing the tail products to some of the vendors out there might seem like a good idea (especially since they include setup and QA), but our research has shown that vendors charge $6.99-$15.00 per description for ~20,000 products at 150 words or less. They may write descriptions faster and keep you from having to staff up for your long tail, but they’re certainly not cheaper. (Our research found that the average cost is $11 per description.)
What about time? Assuming a 2,087-hour work year, 10,000 descriptions will take nearly five person-years to produce if your writers do literally nothing else with their working time (Meetings? Nope. Email? Ha. Training? HAHAHA.). Clearly, you’ll miss quite a few product lifecycles by employing only one human being, so you’ll need ~10 people to hit six months, 20 for three months, etc. Like the monetary investment, the time investment makes sense for your head products, but the investment and overhead for that many human beings starts getting prohibitive fast.
Some retailers have attempted to get around these monumental time and money objects by using the product descriptions that brands provide with the products. This certainly leads to a richer product detail page experience that will help conversions, but lack of unique content at a higher category level doesn’t help as much with SEO as having a unique product description.
Looking at all of this, we realized that we needed to help our customers generate unique product descriptions at scale. But how? We’re a big data company, and scaling paragraphs of text wasn’t in our wheelhouse. We had product information to spare, but no way of bridging the gap to great descriptions.
Enter Narrative Science. They had the opposite problem with solving their customers’ needs for product descriptions: Their advanced natural language generation (NLG) technology could create unique descriptions for any product if they had information about the product itself. Sadly, their customers didn’t necessarily have the product attributes that Narrative Science needed to produce these descriptions.
Indix had product attributes without NLG. Narrative Science had NLG without product attributes. Forming a partnership was, frankly, a no-brainer.
In our post going into more detail about Indix Standardized Attributes, we discussed how we went from the messy “soup” of non-standardized attributes to the organized and vastly more useful collections of standardized attributes, like these for a window box planter:
Narrative Science takes this set of standardized attributes and combines them with a tone document (for brand voice) and a lot of NLG “magic” from their Quill platform and produces an SEO-optimized, customer-friendly product description. This process can produce up to 1,000,000 product descriptions in eight weeks, and is the fraction of the cost of employing human beings to do it (we start at $4/record with volume discounts).
Here are a few more examples of these NLG descriptions produced by Narrative Science from Indix Standardized Attributes:
For more information from Narrative Science and Indix about how we came together to quickly produce cost-efficient, SEO-optimized product descriptions, check out the Narrative Science Automated Descriptions White Paper and sign up for our upcoming webinar on May 17, 2017.
Also published on Medium.